High-Performance Replicated Distributed Objects in PartitionableEnvironments

This paper presents an implementation of replicated distributed objects in asynchronous environments prone to node failures and network partitions. This implementation has several appealing properties: It guarantees that progress will be made whenever a majority of replicas can communicate with each other; it allows minority partitions to continue providing service for idempotent requests; it offers the application the choice between optimistic or safe message delivery. Performance measurements have shown that our implementation incurs low latency and achieves high throughput while providing globally consistent replicated state machine semantics. The paper discusses both the protocols and interfaces to support efficient object replication at the application level